fix(tb-lf): align eval response structs with devportal API

[dafilipaj]

Summary

tb-lf eval run <id> (v0.6.0) failed with invalid type: sequence, expected a string at line 1 column 998, and tb-lf eval cases [--suite <key>] failed with invalid type: map, expected a sequence at line 1 column 0. Both bugs were client-side struct/shape mismatches against DevPortal's actual response.

EvalItem (crates/tb-lf/src/types.rs) was out of sync with SpaApi::Ai::Eval::RunsController#item_json:

• conversation_log: Option<String> — server returns a JSON array of {role, content} (a JSON column, surfaced via as_json). Switched to Option<serde_json::Value>. This is the field that triggered the deserialize failure.
• Renamed suite→suite_key + added suite_name, case→case_key + added case_name, duration_seconds→duration_ms, trace_langfuse_id→trace_id — matching the keys actually emitted by the controller. Previously these silently deserialized to None, leaving the items list rendered as blank rows.
• Display formatting in main.rs updated accordingly; duration is now duration_ms / 1000.0 for the Xs display.

EvalAction::Cases (crates/tb-lf/src/main.rs) was decoding into Vec<EvalCase>, but CoverageController#cases renders { data: [...], meta: {...} }. Switched to PaginatedResponse<EvalCase> and use resp.data. Also fixed the suite filter query param: server expects suite_key, CLI was sending suite, so --suite silently didn't filter.

Added deserialize tests covering both endpoints with representative payloads so the contracts can't drift again silently.

Test plan

Local verification against live DevPortal (Development project):

• cargo run -p tb-lf -- eval run 60 --project Development — renders 352 items with populated suite / case / status / score / duration columns (previously failed at first item with conversation_log)
• cargo run -p tb-lf -- eval run 60 --failed --project Development — filters to failed only
• cargo run -p tb-lf -- eval run 60 --full --project Development — prints truncated JSON conversation log per item
• cargo run -p tb-lf -- eval cases --project Development — lists 50 cases (default per_page) from the paginated wrapper
• cargo run -p tb-lf -- eval cases --suite crm-agent --project Development — filters to 2 CRM Agent cases (previously: did not filter, deserialize-failed before display)
• cargo fmt --check, cargo clippy --workspace -- -D warnings, cargo test --workspace — all clean (5 tb-lf unit tests now, +2 regression tests)

Task

n/a — incidental tooling fix surfaced while running evals.